ovml-training

Use Yolo_mark to mark bounding boxes for training ovml networks.

Credit: this repo is a clone of Yolo_mark, modified/preconfigured for ovml training.

Usage

1. Compile the executable (see below)

2. Assemble your collection of images, which are usually frame grabs from video files. You can use e.g. ovideo::ov_video_frame for this if you have video files locally, or https://www.youtubescreenshot.com/ to extract frames from YouTube videos

3. Put your .jpg images into the data/img directory

4. Run ./mark_ovml.sh (Linux) or mark_ovml.cmd (Windows). You might find it useful to make the bounding box lines thinner (cycle through line widths by pressing w) and hide the object names (k). See below for further key controls or press h

5. Send us the contents of the data/img directory (the image files and their bounding box .txt annotation files)

Notes on images

• Images should generally have all or most of the court in view. For ball detections, the minimum image size (video resolution) should be 1280 x 720

• Try and collect images with varying ball colouration, lighting, court surrounds, and backgrounds

• Avoid images with blurry balls (mild blurring is OK), or balls that are more than about 20% occluded by other objects (player bodies, net posts, etc)

• Mark all target objects in an image. If you are marking balls, don't just mark the ball in play. All other balls in view must also be marked. You might want to avoid images with ball trolleys full of balls, or balls all over the court during warmup for this reason

• The bounding box should be tight around the object but include all of it

Compilation

• To compile on Windows open yolo_mark.sln in MSVS2013/2015, compile it x64 & Release. Change paths in yolo_mark.sln to the OpenCV 2.x/3.x installed on your computer:

  • (right click on project) -> properties -> C/C++ -> General -> Additional Include Directories: C:\opencv_3.0\opencv\build\include;

  • (right click on project) -> properties -> Linker -> General -> Additional Library Directories: C:\opencv_3.0\opencv\build\x64\vc14\lib;

• To compile on Linux:

  cmake .
  make
  

---

Instruction manual

Mouse control

Button	Description
Left	Draw box
Right	Move box

Keyboard Shortcuts

Shortcut	Description
→	Next image
←	Previous image
r	Delete selected box (mouse hovered)
c	Clear all marks on the current image
p	Copy previous mark
o	Track objects
ESC	Close application
n	One object per image
0-9	Object id
m	Show coords
w	Line width
k	Hide object name
h	Help